Client
Strakovo Garden Center is a Czech garden center with an extensive online catalog. Its customers often do not search for a specific cultivar. They need advice based on their conditions and what they want to achieve.
Challenge
Online advice depended on the capacity of qualified staff. Customers asked questions such as “which apple tree is suitable for a small garden,” “which roses grow in shade,” or “which grasses up to 1.5 m are suitable for partial shade.” Answering these questions requires knowledge of the product range, plant characteristics, and the conditions described by the customer.
Strakovo’s management therefore required an assistant that would provide advice exclusively from the company’s own content and current product range. It was not to recommend plants Strakovo did not sell or provide answers outside the information available from the garden center. The goal was to make expert advice available on the website around the clock without expanding the team in proportion to growing inquiry volume.
Solution
Aspecta deployed an AI assistant that works with Strakovo’s product data and educational content. The catalog contains thousands of products with both Latin and Czech names, as well as variants of the same cultivars that differ in attributes such as container size or rootstock. To make this data usable, we adjusted the category structure, combined short and long descriptions, mapped Latin names to Czech names, and processed individual variants. We also connected the garden center’s educational materials.
Technology foundation: the assistant runs on AWS. Amazon Bedrock provides the Claude (Anthropic) language models, answers are based on a RAG layer over the product catalog and documents, and security guardrails protect the assistant against misuse.
The e-commerce XML feed provides the product data. Before launch, Strakovo’s management flagged questions with unsatisfactory answers, and Aspecta used this feedback to adjust the data and tune the assistant.
Results
- In production, the assistant handles approximately 2,500 customer questions per month.
- Approximately five months passed between the first contact in October 2025 and go-live in March 2026.
“I reviewed the questions and answers in detail. So far, they look very good. I estimate that the assistant provides relevant answers in 80 to 90% of cases.” (garden center management, March 2026)
“The assistant has to be reliable. It is the ‘flagship’ of our communication.” (garden center management, April 2026)
Post-launch process
The XML feed synchronizes daily, and the solution continues to be evaluated using real customer questions. The client receives regular reports showing which questions the assistant handled well, where an answer was not ideal, and what to refine in the XML feed or add to the content. Aspecta continues to adjust the solution based on the client’s feedback. The core rule remains unchanged: the assistant answers only from Strakovo’s content and product range.
Another example from the AI smart assistant series: a Slovak e-commerce store for paints and chemicals, Farby.sk.
